Meta is working to enhance the privacy of users on its AI platform. The tech giant is collaborating with Moxie Marlinspike, known for Signal and its open-source encryption protocol, to bring the technology from his privacy-focused AI platform Confer into its system. Marlinspike made the announcement in a blog post.
Until now, end-to-end encryption technologies have largely been limited to instant messaging platforms such as WhatsApp and Telegram. For Signal, too, it is a big deal. These systems ensure that messages remain visible only to the sender and the receiver, preventing even the platform itself from accessing them.
Growing concerns around AI chatbot privacy
As the capabilities and popularity of AI chatbots continue to grow, users are exchanging billions of messages with these systems, often including confidential information. In his blog post, Moxie Marlinspike said that large language models (LLMs) are impressive but also represent the “largest centralised data lakes in history, containing more sensitive data than anything ever before.”
AI companies rely on such data to train their models, and many platforms have made it difficult for users to opt out of having their information used for training. Marlinspike said, “We are using LLMs for the kind of unfiltered thinking that we might do in a private journal – except this journal is an API endpoint to a data pipeline specifically designed for extracting meaning and context.”
He also added that “right now, none of that data is private.” The absence of end-to-end encryption exposes conversations with AI to companies, employees, hackers and governments, raising concerns that such information could end up in the wrong hands. As these platforms continue to grow in popularity, even more personal data from users is expected to flow into these systems.
With the rising use of AI chatbots, technologists and privacy advocates are increasingly calling for platforms to implement stronger privacy protections. Marlinspike said, “Confer is built so that nobody has access to your conversations but you (not even me!).”
However, applying encryption to AI systems is not straightforward. The end-to-end encryption used in traditional digital communication cannot easily be transferred directly to generative AI chatbots. Confer’s technology is built on open-source models, and its collaboration with Meta will provide an opportunity to test it alongside closed AI models.
Marlinspike’s collaboration with Meta is not new. He previously partnered with Meta-owned WhatsApp in 2016 to roll out end-to-end encryption on the messaging platform.
Encryption debate continues
However, the development also highlights a contradiction in Meta’s broader approach to encryption. The company has confirmed that it will remove end-to-end encryption for Instagram direct messages after May 8, 2026. Meta has cited concerns about child sexual abuse material (CSAM) on the internet as the reason for the move.
According to the company, removing encryption will allow it to scan direct messages and calls on Instagram to detect content related to CSAM, grooming and other forms of harassment.
